That is an excerpt of Sources by Alex Heath, a e-newsletter about AI and the tech trade, syndicated only for The Verge subscribers as soon as per week.
I spent yesterday at Eric Newcomer’s Cerebral Valley convention in San Francisco, which is now in its third yr. I’ve attended this occasion for 3 years in a row as a result of Eric does a terrific job curating the audio system and the viewers, and the conversations are extra substantive than a typical trade occasion.
This yr was no exception; nevertheless, I discovered essentially the most fascinating a part of the day to be when the outcomes of an nameless viewers survey have been shared onstage. The greater than 300 attendees who participated within the survey primarily consisted of AI firm founders, adopted by buyers, different trade professionals (together with product leaders and engineers), and members of the media.
Listed here are the outcomes of the survey so as of how they have been shared onstage:
1. What might be OpenAI’s annualized income be on the finish of 2026?
Median reply: $30 billion.
2. What’s going to Nvidia be value on the finish of 2026?
Median reply: $6 trillion.
3. What yr will an impartial committee of specialists, as dictated by the Microsoft-OpenAI settlement, declare that now we have reached AGI?
4. Which enterprise capital agency’s AI portfolio are you essentially the most jealous of?
The highest three most voted for, from first to final: Andreessen Horowitz, Khosla Ventures, and Sequoia.
5. Should you might put cash in any personal know-how corporations at the moment, what would they be?
Prime 5 corporations so as from first to final: Anthropic, OpenAI, Cursor, Anduril, SpaceX, and OpenEvidence.
6. What world firm’s mannequin will high the LMArena net growth leaderboard on the finish of 2026?
So as from first to final: OpenAI, Anthropic, Gemini, Grok, Qwen.
7. Should you might quick a $1 billion-plus valuation startup, which would it not be?
First place was Perplexity. Second place went to OpenAI. Different names proven onstage: Cursor, Determine, Harvey, Mercor, Mistral, and Considering Machines.
What stood out to me from these outcomes (Newcomer has revealed the slides for his paying subscribers):
- A softening on OpenAI: Provided that Sam Altman has mentioned OpenAI plans to finish this yr with $20 billion of annualized income, this group of AI insiders doesn’t anticipate subsequent yr to be as exponential for the enterprise because the leap from 2024 to 2025. The prediction that AGI received’t be declared till 2030 suggests a scarcity of religion in mannequin progress meaningfully bettering within the close to time period, though that reply is also clouded by the complexity of how OpenAI and Microsoft should choose the way it’s determined. (I’m nonetheless ready for both firm to share data on who its “impartial committee of specialists” might be and the way they’ll resolve.) It was additionally notable that extra attendees needed to purchase Anthropic inventory than OpenAI’s, regardless of the consensus being that OpenAI would lead LMArena subsequent yr.
- Meta wasn’t within the dialog. It wasn’t named on the record of fashions more likely to lead LMArena subsequent yr. The presence of a Chinese language mannequin (Alibaba’s Qwen) within the high 5 alerts a shift that’s already underway, as many corporations fine-tune open-source Chinese language fashions somewhat than Llama. Meta has lots to show if it desires to re-enter the mannequin race.
- Perplexity is controversial. However everybody working in AI already is aware of that.
Different takeaways from Cerebral Valley:
What’s driving reverse acquihires? I attended a breakout session about AI acquihires, similar to Meta’s take care of ScaleAI to rent Alexandr Wang and Google’s offers with Character and Windsurf. I’ve intently lined many of those offers over the previous couple of years, nevertheless it was fascinating to listen to the group’s perspective on what drives them. Antitrust scrutiny of Huge Tech actually performs an element, however some who’ve been concerned in these sorts of transactions additionally made the purpose that larger corporations are racing one another to shore up expertise and transfer quicker than their competitors. They’ve seemingly “infinite cash,” as one member of the group put it, and see it as a recreation of inserting bets on a really finite pool of expertise. One AI founder within the group, who fielded a number of gives of this sort, recalled a member of a Huge Tech firm’s company growth crew asking him how a lot he needed his startup to be valued for a deal.
Nobody cares about AGI anymore. On the first Cerebral Valley convention, the subject of AGI was a serious throughline. A startup founder onstage mentioned that “we’re going to be useless” by the point OpenAI releases GPT-10. This yr, a number of onstage conversations famous how AGI barely registered as a dialogue subject. As a substitute, a lot of the interviews targeted on the enterprise functions of AI. A number of corporations represented onstage on the first Cerebral Valley occasion didn’t exist and are actually value billions of {dollars}. There was a pressure of AI bubble concern all through the day, however principally, everybody appeared dialed in on how they might win market share and supply merchandise that folks wish to pay for.
Standout quotes from onstage interviews:
- Replit CEO Amjad Masad: “If you’re competing on value, then perhaps you don’t have a enterprise.”
- Elad Gill: “Most corporations ought to promote sooner or later. There’s usually a market-maximizing second the place you’re going to get the perfect deal you may. A really small variety of corporations ought to by no means ever promote.”
- San Francisco Mayor Daniel Lurie: “Persons are beginning to complain about site visitors. Thank goodness. I would like these complaints. We nonetheless have lots of empty workplace area.”
- Anthropic CPO Mike Krieger: “Time spent is, I can inform you, not on any of the dashboards that I have a look at. It’s simply not a essential consideration.”
- xAI co-founder Jimmy Ba: “Information is simply crystalized computation from the previous.”
